Curriculum design

35 Objectives Ictal-interictal subtraction SPECT with co-registration to MRI provides a valuable non-invasive method to locate seizure-onset zone. Both conventional subtraction methods (e.g. SISCOM) and statistical parametric mapping (SPM)-based methods (e.g. STATISCOM) have been employed to aid epilepsy pre-surgical work-up. Among them, SPM methods accounting for normal variation between two SPECT scans have been shown to provide better localization. However, SPM-based methods require specialized programming tools and are more complex to use than subtraction methods. Commercial software packages that are more easily integrated into the clinical practice are now becoming available. The current study compares SISCOM, STATISCOM, and a commercial package, MIMneuro (MIM Software), which uses both cluster analysis and statistical differences in co-registered images, in epilepsy patients. Methods We retrospectively reviewed 378 patients who underwent ictal/interictal Tc99m-ECD SPECT scans. 28 patients were identified who had SPECT and MRI data sufficient for image analysis, surgical resection of the suspected epileptogenic site and follow-up for at least one year. Hyper- and hypo-perfusion SPECT images were analyzed and co-registered to MR images using SISCOM, STATISCOM and MIMneuro. SISCOM analysis was performed with a z-score of 1.5 and 2, respectively. STATISCOM tool compared difference images with normal controls using an unpaired 2-sample t-test. MIMneuro was performed with a camera resolution of 7 mm and no additional smoothing. A blind review tool was built to randomly present a set of co-registered hyper- and hypo-perfusion images in the same coordinates and color scales. Two observers, blinded to all clinical information except radiotracer injection timing and brief scalp EEG findings, marked up to 2 regions of the most prominent hyper- and hypo-perfusion with confidence ratings. Resection site was used as a reference to calculate the sensitivity of each analysis method. Results The averaged sensitivity of STATISCOM was 85.7%, followed by MIMneuro (83.9%), both of which were superior to SISCOM with a z-score of 2 (67.9%) and 1 (51.8%). When based on the observers’ first choice of seizure localization, all sensitivities decreased with STATISCOM and MIMneuro remaining the highest (both 75.0%). MIMneuro yielded the best inter-observer agreement between the two observers (kappa=0.816), while kappa score was 0.742 for STATISCOM, 0.517 for SISCOM with a z-score of 2, and 0.441 for SISCOM with a z-score of 1.5. The averaged confidence rating was 2.5, 2.3, 1.6 and 1.1 for STATISCOM, MIMneuro, SISCOM with a z-score of 2 and 1.5, respectively. In addition, on a standard windows 7 workstation (Intel Core i5-4300U CPU and 8GB of RAM), processing time per patient took about 3-4 min with MIM Neuro, and took 10-15 min with STATISCOM. Conclusions STATISCOM and MIMneuro showed better localization and higher inter-observer agreement compared to SISCOM. MIMneuro took less computation time than STATISCOM and provided further benefit to clinical workflow.